Re: [PATCH] drm/gud: NUL-terminate TV mode names read from the device

Ruben Wauters <[email protected]> Mon, 17 Aug 2026 16:36:09 +0100
Newsgroups	org.freedesktop.lists.dri-devel,org.kernel.vger.linux-kernel
Message-ID	<[email protected]>
On Sun, 2026-08-16 at 09:00 +0000, [email protected] wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> 
> Pre-existing issues:
> - [High] A malicious or misconfigured USB device can cause a NULL pointer dereference and kernel panic in `drm_object_attach_property` by returning `GUD_PROPERTY_TV_MODE` after any other non-margin TV property (such as `GUD_PROPERTY_TV_BRIGHTNESS` or `GUD_PROPERTY_TV_HUE`).
> --
> 
> --- Patch [1]: [PATCH] drm/gud: NUL-terminate TV mode names read from the device ---
> commit 789868bcbf6cc6ddf0d3b74feab3a22914863ec8
> Author: Deepanshu Kartikey <[email protected]>
> 
> drm/gud: NUL-terminate TV mode names read from the device
> 
> This commit fixes a potential out-of-bounds read by explicitly NUL-terminating
> TV mode names read from the USB device. This ensures strlen() won't run past
> the end of the allocation when passed to drm_mode_create_tv_properties_legacy().
> 
> Fixes: 40e1a70b4aed ("drm: Add GUD USB Display driver")
> Reported-by: [email protected]
> Closes: https://syzkaller.appspot.com/bug?extid=916c888ba5f1a54c9526
> Tested-by: [email protected]
> Signed-off-by: Deepanshu Kartikey <[email protected]>
> 
> > diff --git a/drivers/gpu/drm/gud/gud_connector.c b/drivers/gpu/drm/gud/gud_connector.c
> > index ea0cca58b7c89..5c0065c876a77 100644
> > --- a/drivers/gpu/drm/gud/gud_connector.c
> > +++ b/drivers/gpu/drm/gud/gud_connector.c
> > @@ -396,8 +396,13 @@ static int gud_connector_add_tv_mode(struct gud_device *gdrm, struct drm_connect
> >  	}
> >  
> >  	num_modes = ret / GUD_CONNECTOR_TV_MODE_NAME_LEN;
> > -	for (i = 0; i < num_modes; i++)
> > -		modes[i] = &buf[i * GUD_CONNECTOR_TV_MODE_NAME_LEN];
> > +	for (i = 0; i < num_modes; i++) {
> > +		char *mode = &buf[i * GUD_CONNECTOR_TV_MODE_NAME_LEN];
> > +
> > +		/* The device is not trusted to NUL-terminate the name */
> > +		mode[GUD_CONNECTOR_TV_MODE_NAME_LEN - 1] = '\0';
> > +		modes[i] = mode;
> > +	}
> >  
> >  	ret = drm_mode_create_tv_properties_legacy(connector->dev, num_modes, modes);
> 
> [Severity: High]
> This is a pre-existing issue, but could a malicious or misconfigured USB device
> cause a NULL pointer dereference and kernel panic by returning
> GUD_PROPERTY_TV_MODE after another non-margin TV property?
> 
> In gud_connector_add_properties(), properties from the USB device are processed
> sequentially. If a property like GUD_PROPERTY_TV_BRIGHTNESS is processed first,
> drm_mode_create_tv_properties_legacy() creates the baseline TV properties but
> leaves legacy_tv_mode_property as NULL since num_modes is 0.
> 
> If GUD_PROPERTY_TV_MODE is processed next, it calls gud_connector_add_tv_mode()
> which then calls drm_mode_create_tv_properties_legacy() here:
> 
> drivers/gpu/drm/drm_connector.c:drm_mode_create_tv_properties_legacy() {
>     ...
> 	if (dev->mode_config.tv_select_subconnector_property)
> 		return 0;
>     ...
> }
> 
> This early return prevents dev->mode_config.legacy_tv_mode_property from
> being created.
> 

Hello
> Then in gud_connector_add_properties(), gud_connector_property_lookup()
> will return NULL for legacy_tv_mode_property:
> 
> drivers/gpu/drm/gud/gud_connector.c:gud_connector_add_properties() {
>     ...
> 		property = gud_connector_property_lookup(connector, prop);
> 		if (drm_WARN_ON(drm, IS_ERR(property)))
> 			continue;
>     ...
> }
> 
It might be an idea to add a null check here, however, I am inclined to
accept this patch without it, as it does fix an issue with the current
code, and while this issue is related, it is somewhat separate. It
would be a good idea to have this as a separate patch instead.

Ruben Wauters

> Since IS_ERR() ignores NULL, the NULL property pointer bypasses this check and
> is passed to drm_object_attach_property(), which unconditionally dereferences
> it:
> 
> drivers/gpu/drm/drm_mode_object.c:drm_object_attach_property() {
>     ...
> 	struct drm_device *dev = property->dev;
>     ...
> }
> 
> Could this lead to a local denial of service?
> 
> >  free:
> >  	kfree(buf);
> >  	if (ret < 0)
signature.asc (application/pgp-signature, 870 B)